Who runs this, and by what rules

Why trust a site run by one person

You should not have to. The site is built so its claims can be checked without trusting its author. Every number traces to resolved on-chain events anyone can verify. The methodology describes how the scoring works and states its limitations, the alert record grades itself in public with losses included, and the validation report was committed to publication before its results existed. The receipts ledger’s won, lost, voided and sealed counts add up to its total.

Corrections

Errors are corrected the same way alerts are graded: in public. When a published number or graded row is found to be wrong, the correction is applied and noted on the affected page rather than silently swapped, and anything material enough to change a conclusion is called out where the conclusion was drawn. Grading itself is mechanical: alert rows are recorded at detection, frozen, and settled by each market’s recorded resolution, with no editing after the fact. Voided markets are excluded from the record and labeled as such. Every correction is logged below.

Correction log

Every published number, verdict, or claim found to be wrong, newest first, including the ones nobody outside would have noticed. A product that grades other people’s records in public should be able to show its own.

• 2026-10-05: Alert detection skipped buys after a market's listed end time, which Polymarket sets before most esports matches and many US evening games are over. The 2026-09-22 entry said the alert pipeline has never excluded in-play sports markets, and the methodology page says they are included. From launch in July 2026 until 12:52 UTC on 2026-10-05, alert detection skipped any buy in a market whose listed end time had passed when its 10-minute pass ran. Polymarket lists most esports matches as ending at midnight UTC at the start of match day, and many US evening games as ending at midnight UTC (8 PM Eastern) while they are still being played. So in those esports matches no match-day buy (UTC) could alert, and in those US games no buy after 8 PM Eastern could. Daily copies of the alert watchlist (the wallet and category pairs whose trades can alert) have been kept only since 2026-09-27, so a replay covered 2026-09-27 to 2026-10-04. It found 48 positions in markets still trading after their listed end that were not alerted and would have cleared the conviction gate, which needs $1,000 or more, a price between 10 and 90 cents, and at least 1.5 times the wallet's median position over the past 120 days. Medians were measured as of 2026-10-05, and the count leaves out any buy the one-side rule would have stopped. Of the 48, 47 were in Esports and 1 in Sports, 29 were in-play and 19 were match-day buys before the game, and they came from 7 wallets. How many alerts this blocked before 2026-09-27 is not known. Since 12:52 UTC on 2026-10-05, a market that Polymarket gives a game start time no more than two days after its listed end time can alert until the later of that listed end and 8 hours after the game starts. Other markets keep their listed end time. Alerts already sent, the receipts ledger and its record, the scoring method and the boards did not change.
• 2026-10-04: Boards published at 10:27 UTC on 2026-10-04 used the wrong categories. Polymarket changed the order in which it lists each market's tags, starting on 2026-10-03. OVERROUND filed each market under its first tag, so the nightly job that copies market details from Polymarket (04:30 to 05:39 UTC on 2026-10-04) changed the category of 1,155,277 markets. Most were closed crypto markets left without a category, and others moved from Crypto, Sports and Politics into narrower tags such as Bitcoin, Games and Midterms. The boards published at 10:27 UTC were built from those categories, with 471,509 wallet entries against 544,801 on 2026-10-02. Some boards lost markets, others gained them, and boards appeared for tags such as Games and Bitcoin. Wallet pages showed per-category histories and calibration from that build until 17:39 UTC. Categories were restored from the backup taken before the nightly job. Markets it already held mis-filed, and markets created after it, were re-filed by a rule that picks the category from all of a market's tags in any order, which matched the earlier categories on 99.54% of the markets it was tested on. The same rule now files new markets (the 2026-09-25 entry describes the old one), and the nightly job no longer replaces a stored category. The boards were republished at 17:39 UTC with 546,589 wallet entries. Because 2,807 closed markets were re-filed only at 18:26 UTC, after the republish, they stay under their wrong categories on the boards until the nightly update on 2026-10-05. The wallet checker, which pools a wallet's record across categories, and the scoring method were unaffected.
• 2026-10-04: An alert watchlist built from the wrong 2026-10-04 boards was used for an hour, and watched wallets' buys in mis-filed markets could not alert. The alert watchlist (the wallet and category pairs whose trades can alert) rebuilt from the wrong boards at 11:06 UTC on 2026-10-04 (see the entry above) had 4,545 pairs across 307 categories, against 1,310 across 71 the day before, until the previous day's was restored at 12:04 UTC. No alert was sent from it, and wallets it dropped could not alert during that hour. Alerts match a market's category, so a watched wallet's buy in a mis-filed market could not alert under the category the wallet is watched for, from 2026-10-03 for markets created that day and from 04:30 UTC on 2026-10-04 for the rest, until most were re-filed at 12:11 UTC and the last open ones at 14:25 UTC. Watched wallets bought $50 or more in 11 positions in mis-filed markets or while dropped from the watchlist, and none could have become an alert (an alert needs $1,000 or more at a price between 10 and 90 cents, among other conditions), so no alert to subscribers was missed. The watchlist rebuilt from the republished boards at 17:44 UTC has 1,309 pairs, 1,296 of them on the previous day's list, and the 13 new ones made no buy of $50 or more between 11:06 and 17:44 UTC. Alerts already sent, the receipts ledger and its record were unaffected.
• 2026-10-03: Calibration benchmark: a sentence said politics scores worse than the platform average, and the page's 24-hour table gives politics the lower Brier score. From 2026-07-23 until 2026-10-03, the section of the 2026 Q3 calibration benchmark on why public accuracy figures disagree said political markets price closer to 50 cents and score worse than the platform average at every horizon. The page's tables contradict both halves. A day before resolution, its Politics row has a Brier score of 0.0690 and 90.1% accuracy, against 0.1253 and 81.2% for the full 24-hour universe (a lower Brier score is better). Almost two in three of those politics markets were priced under 10 cents, against about two in five across the full 24-hour universe. The page gives Brier scores by category only at the 24-hour horizon, so nothing on it supported "every horizon". The paragraph now states the 24-hour comparison, says that markets with a heavy favorite score well by construction, and adds that politics markets between 10 and 50 cents missed their prices by more than markets in the same range across the full 24-hour universe did. No figure on the page changed.
• 2026-09-29: A hand-kept list the site said muted market-maker wallets from alerts was empty. On 2026-06-13 we added 10 wallets, judged in earlier reviews to be market makers or bots, to the list that hides wallets from the public board. That list does not touch alerts, and the separate list that mutes a wallet from alerts was empty. On 2026-09-29, seven of the 10 held 19 places on the alert watchlist, which is rebuilt from the scores every night. Between 2026-07-14 and 2026-09-24, 36 alerts on their positions were sent. Six of them reached a paying subscriber, between 2026-09-11 and 2026-09-24, as well as our own test accounts. The other 30 reached only our own test accounts. From 2026-09-27 until 2026-09-29, the methodology page, the alerts page and the 2026-09-27 entry on alert screens said alerts leave out a hand-kept list of wallets muted as market makers, and that list muted no one. Before then the site said market makers were screened out, as that entry describes. All 10 wallets were muted from alerts on 2026-09-29. Their places on the watchlist are left empty, and no lower-ranked wallet moves up. The alerts already sent stay on the receipts ledger and in its record. Scores and grades did not change, and the July validation had already left these wallets out.
• 2026-09-29: A research label named for a copier's price, an alert warning a test did not bear out, and a scoring reason a check did not support. The research label on the board, wallet pages and checker said whether a record held up at a copier's price. That price is the first trade at least five minutes after each fill, whichever side started it, and it is not what a copier pays. In a pre-registered run on 2026-09-28, at the taker price (the first purchase at the asking price, or for a sale the first sale at the bid, five or more minutes after each fill), wallet-categories whose earlier records held up came out 0.21 cents a share below the price curve on later events, before fees. The label is now named for the 5-minute price, and its note gives that result. From 2026-09-27, an alert from a wallet whose record did not hold up carried a warning line. A pre-registered test on 2026-09-29 did not show those records doing worse later than the rest, so the line was removed. The high-price flag in alerts said the edge was mostly captured, which implied an edge remained below 90 cents. It now states the price and how much is left to win. The methodology page gave, as the reason the price baseline is zero between 30 and 70 cents, that returns there are the footprint of informed traders, and said two of its measurements were findings we would publish separately. A pre-registered check did not support that reading, and the page now says so. It also called the alerts fewer and better, and the word better is gone. The copy-trading guide reported only the test at the first trade five minutes later. It now reports the taker-price run beside it, and a line saying a ranking by accuracy describes what a copier inherits is removed. The homepage said scores use a small set of proprietary anti-gaming screens. No such screens exist, and every scoring rule is on the methodology page. Scores, grades and the published test results did not change.
• 2026-09-27: Methodology and research: alert screens that did not run, rank stability, and what the validation's alert slice covered. The methodology page said a wallet passes structural and behavioral screens before it generates alerts, and that each misleading pattern is removed before a wallet is ranked. Besides the score rules, the conviction gate and the one-side rule, the only filter on wallets is a hand-kept list of wallets muted as market makers or otherwise impossible to follow, or delisted at their owner's request, and a wallet new to the alert watchlist is not reviewed, by hand or by any automatic market-maker test, before its first alert. Two guides described market makers as screened out and the alerts page called selection mechanical; all three now say the list is kept by hand. The methodology page also said our top wallets stay on top under every reasonable variant of the scoring we tested; the v1.1 and v1.2 updates moved the leader on some boards, so the line overstated how stable the rankings are. The research page called the validation's alert-eligible group the slice the paid alerts draw from, and the statistics page used alert-eligible without saying what it covered. That group is the wallets that met the watchlist's score rules at each date, before its limit of 25 per category and without its hand-kept exclude list (it did leave out wallets hidden from the public board), and the July report did not test the watched group on its own. The statistics page now also says its calibration table measures wallets on the same trades they were ranked on. All of these were corrected on 2026-09-27. No alert, score, grade or published figure changed.
• 2026-09-27: Two guide lines still reflected win-rate figures from before the 2026-09-24 refresh. When the win-rate figures were refreshed on 2026-09-24 (scoring v1.1), two lines on other guides were missed. The most-accurate-traders guide kept the 2026-09-22 median win rate for wallets with 1,000 or more graded events, 57.7%, where the 2026-09-24 run gave 61.3%. The leaderboard guide still said the median wallet in the thinner evidence tiers performs slightly below its own entry prices, but on the 2026-09-24 run the median record with 20 to 99 events sat 0.8 wins per hundred ahead of them, while the 100 to 999 tier's sat 0.1 behind, so the line held for one of the two tiers. Both lines were corrected on 2026-09-27, and the most-accurate-traders guide's median now reads 61.0% from that day's run under the corrected checker test. No verdict, ranking or alert was affected.
• 2026-09-26: Wallet checker: whether a position on several candidates in one race counted depended on who won. The 2026-07-25 hedged-positions correction kept hedges and early exits out of the checker by leaving out any event unless every bet the wallet made in it won, or every one lost (a sale counts as a bet on the other side). Across several markets of one event, that test depended on which outcome won. A wallet holding No on two candidates in a three-way race was counted when the third candidate won and left out when either candidate it bet against won, so the checker counted that strategy's wins and dropped its losses. Yes on several candidates had its losses counted and its wins dropped. A record built on such positions could read as ahead of or behind its prices because of which events the test kept. From the 2026-09-27 nightly update the checker counts only positions on a single market held on one side. That test reads only the wallet's own trades, and positions on several markets we count as one event are left out whatever the result. Verdicts and evidence ranges changed for some wallets that bet on several markets of one event, such as several candidates in one race. Shared checker results made before then keep the verdict and date they were made with. The win-rate figures quoted on the win-rate, copy-trading, skilled-or-lucky and most-accurate-traders guides come from the same test and were refreshed from the 2026-09-27 run. The leaderboard, alerts and graded histories never filtered on either test and are unaffected, and so are alert grades on the receipts ledger.
• 2026-09-25: Scoring v1.2: the Sports board from the 2026-09-25 update carried no ranking. From the 2026-09-25 nightly update until the 2026-09-26 one, every wallet on the Sports board got the same score, the category's baseline, so the board's order carried no ranking information and no alert could fire on a market filed under Sports. The Fed board got the same treatment in the 2026-09-25 update. The cause was the evidence weighting. It estimates how much real skill varies within a category as the spread of the category's records minus the spread luck alone would produce, and that luck estimate treated every wallet in a category as equally noisy, using one pooled noise level weighted toward the wallets with the most events. In Sports that level ran higher than the average wallet's own noise, and the estimate came out at zero. The problem first showed once tennis and esports markets returned to their own boards (the next entry). Methodology v1.2, applied from the 2026-09-26 nightly update, puts each wallet's own noise into the luck estimate. The Sports board and its alert watchlist are rebuilt with that update, and the change is smaller elsewhere. The Fed board still comes out at zero under v1.2, so its order carries no ranking information. The statistics page's calibration figures for top-ranked wallets were refreshed on 2026-09-26 under v1.2 and the corrected tennis and esports categories. On their trades in the 20 to 80 cent range, the figure moved from about 3.4 to about 3.0 points better than the price paid. The July validation results were computed under v1.0 and stand as published, and alert grades on the receipts ledger are unaffected.
• 2026-09-25: Tennis and esports markets were filed under Sports in the 2026-09-24 update. On 2026-09-23 Polymarket put Sports first in the tags of tennis and esports markets. OVERROUND files a market under its first tag, so the 2026-09-24 nightly update scored most tennis and esports history inside Sports. That day's Tennis and Esports boards held only a small remainder of their records, the Sports board carried their wallets, and alerts on tennis and esports markets mostly stopped. A fix on 2026-09-25 keeps a Sports-first market under Tennis or Esports when it carries one of those tags, and the 2026-09-25 update rebuilt both boards and their alert watchlists. The statistics page's calibration figures dated 2026-09-24 and the follower-price test published on the research page on 2026-09-24 were computed on the mis-filed data (the calibration figures were refreshed on 2026-09-26; see the entry above). The test was rerun with the same scripts on the corrected categories on 2026-09-25, and the rerun, which decides the result, is published beside the first run on the research page. In the rerun the evidence test passed, where the first run had failed, and the volume check fell short in both. Alert grades on the receipts ledger are unaffected.
• 2026-09-24: Scoring v1.1: bets at 2.5 cents or less no longer count, and the win-rate figures moved. Some category leaders had built their records mostly from buys at a few cents. The scoring weighs every share in an event alike, so a few dollars of penny buys could account for most of a record's score, and OVERROUND alerts never go out at those prices. From 2026-09-24 (methodology v1.1), bets at 2.5 cents or less are left out of scoring, such as a buy at 1 cent, or a sale at 99 cents, which bets on the other side at 1 cent. The boards, the alert watchlist and the wallet checker changed with the 2026-09-24 nightly update. The win-rate figures on the win-rate and copy-trading guides moved too. An event that held a favorite beside a penny bet on the other side now grades as a one-sided favorite bet, so more wallets reach the 20 graded events the guides count, and between the 2026-09-21 and 2026-09-24 runs the graded population rose from 402,094 wallets to 484,465. In the 20 to 99 event tier the median record now runs 0.8 wins per hundred ahead of its own prices, where the v1.0 figure was 0.9 behind, and the win-rate guide's conclusion that the median wallet loses to its own prices is withdrawn. The share of records behind their prices at 95% confidence fell from 12.2% to 3.6% at 20 to 99 events and from 30.7% to 9.4% at 100 to 999. The statistics page's calibration figure for top-ranked wallets, measured on trades in the 20 to 80 cent range, moved from about 4.0 to about 3.4 points. The July validation results were computed under v1.0 and stand as published. Alert grades on the receipts ledger are unaffected.
• 2026-09-22: Competitor pages: claims re-checked and rewritten. The Polyburg review and the alternatives list on the alternatives pages carried claims checked on 2026-07-22 and still labelled verified after they had gone out of date. The review described an alert terminal, a published win-rate selection formula, a case-studies page of winning alerts and no free trial. Today Polyburg's main product is copy trading, the formula page returns a 404, the case-studies page lists no entries, and the first month of membership is free with an invite. Shadower's prices had changed, Convexly now sells wallet-move alerts, and the Unusual Whales entry called the product positioning-based, though its page advertises category-specific trader scoring and historical accuracy tracking. Two claims were wrong when published. Arkham's announcement puts its top trader by P&L in the top 7% of Elo ratings, and our review called that rating mid-table. The OVERROUND entry said none of the listed tools carried a pre-registered validation or a graded record of its own output, and Convexly publishes both. Every competitor claim was re-checked in a browser on 2026-09-22 and the pages rewritten. Claims that depended on Arkham's live leaderboard, which would not load, were removed. OVERROUND's own figures are unaffected.
• 2026-09-22: Methodology: the alert exclusions, criteria and cadence described what the pipeline does. The methodology page said alerts exclude markets that resolve too fast to act on. The alert pipeline has never excluded in-play sports markets; it excludes the Up or Down series, daily highest-temperature markets and above/below price-threshold markets, and the page now says so. It also listed timing among the alert criteria, and no timing criterion exists. Its cadence line, three to five alerts a day, undersold busy sports days and now reads a handful a day, more on heavy sports days. No alert, grade or ledger figure changed. The same day, the statistics page dropped its table of named profit-board accounts graded one by one, and its alert record and coverage figures now read live from the public API.
• 2026-08-05: Receipts ledger: re-alert snapshots were counted as separate rows. Detection re-alerts a position as the tracked wallet adds to it, and each snapshot carries the position's cumulative size. The ledger recorded every snapshot as its own row, so one growing position could appear several times: 1,430 rows over 644 distinct positions. Re-alerted positions won more often than average, so row counting flattered the record: 723 won to 379 lost by rows (66% of resolved), 344 won to 209 lost by positions (62%). The ledger now carries one row per position, keeping the first alert's date, price, and the market prices 1 and 24 hours after it, the largest size observed, and an n_alerts column in the public JSON ledger stating how many times the feed flagged the position. Individual grades were never wrong; every snapshot of a position grades identically. The same double counting inflated the daily email and Telegram digest's header totals, fixed the same day. Figures are from the 2026-08-05 verification run.
• 2026-07-26: Win-rate guide: skill-prevalence shares replaced after the hedging correction. The guide published the share of records statistically distinguishable from luck as 6%, 27%, and 45% by evidence depth. Two things were wrong with that. The test behind it counted hedged and traded-out events as forecasts, which inflates thin records, and the shares were computed as the ahead-of-prices tail alone while the wording did not say so. On the corrected test, which grades only positions held to resolution on one side, the same ahead-tail statistic reads 1.3%, 4.5%, and 44.5%, and the page now states both tails: 12.0%, 29.8%, and 22.4% of records are distinguishable and behind their prices. The graded population moved from 640,363 to 377,736 wallets. Our interim note predicted the figures would fall substantially; that held for the thin tiers, and the deepest tier barely moved. Every figure on the page comes from the 2026-07-26 verified run, and distinguishable shares are stated by direction everywhere they appear.
• 2026-07-25: Wallet checker: hedged positions were graded as forecasts. The checker's test treated each event as a single bet at the price paid. Where a wallet held both sides of an event, or entered and exited before it resolved, that assumption was wrong, and wrong in one direction. A pair bought at 40 cents a side profits whichever way the event lands, so arbitrage and profit-taking were scored as forecasting accuracy. 39% of all wallet-events carried that signature. The checker now grades only positions held to resolution on one side and says so on every result. The skill-prevalence shares on the win-rate guide came from the same test and are marked under revision until the next verified data run replaces them.
• 2026-07-25: Wallet checker: a record behind its own prices could read as clearing the bar. The significance test flags a record as distinguishable in either direction, better or worse than the prices it paid. Every surface rendered that as a positive result, and the checker attached a subscription prompt to it. Wording and next steps are now split by direction, and a record that resolved behind its prices never receives the positive framing.
• 2026-07-25: Wallet checker data was four days old behind a current timestamp. The table behind the checker was published to the live site but was never included in the nightly rebuild, so it carried figures from 21 July while the page reported a current as-of date. It is rebuilt nightly now, and an automated check fails the build if any published table is missing from the rebuild.
• 2026-07-25: Statistics page figures were ten days old. The graded-alert counts read 745 sent and 517 resolved while the live ledger showed 1,043 and 726. Each block on that page now carries the date of the data run that produced it, so one stale section cannot sit behind a page-level date.
• 2026-07-24: Benchmark: a claim about political markets was not supported by the data. The calibration benchmark said our politics figures showed the price compression reported in the academic literature. A category average cannot demonstrate that, and the underlying price-bin curve ran the other way at short horizons. The claim was replaced with what the data shows, and the supporting table was published alongside it.

Methodology changes

The scoring method is versioned and dated on the methodology page. When the method changes in a way that moves rankings, the change is recorded there with its date and reasoning; historical grades on the receipts ledger are never rescored retroactively.

Contact

Questions, corrections, or data requests: hello@overround.pro. Data journalists and researchers can quote the statistics page with attribution; the public datasets are licensed CC BY 4.0.